# How to Teach Large Multimodal Models New Skills

arXiv: https://arxiv.org/abs/2510.08564 · HTML: https://mlsanity.com/paper/2510.08564/

## Key facts

- Cited by 11 papers in the map (data through 14 Oct 2026).
- Placed in Prompt learning and adaptation of CLIP, one of 16 directions in Multimodal LLMs / vision-language models.
- Builds on 60 of its 77 references that are in the map.
- First posted on arXiv 9 Oct 2025.

## Placed in

- Multimodal LLMs / vision-language models / [Prompt learning and adaptation of CLIP](https://mlsanity.com/mllm/prompt-learning-and-adaptation-of-clip/)

## Builds on (most cited first, 30 of 60)

- [Learning Transferable Visual Models From Natural Language Supervision](https://mlsanity.com/paper/2103.00020/) (2021)
- [Visual Instruction Tuning](https://mlsanity.com/paper/2304.08485/) (2023)
- [Overcoming catastrophic forgetting in neural networks](https://mlsanity.com/paper/1612.00796/) (2016)
- [LLaVA-OneVision: Easy Visual Task Transfer](https://mlsanity.com/paper/2408.03326/) (2024)
- [Locating and Editing Factual Associations in GPT](https://mlsanity.com/paper/2202.05262/) (2022)
- [Learning without Forgetting](https://mlsanity.com/paper/1606.09282/) (2016)
- [iCaRL: Incremental Classifier and Representation Learning](https://mlsanity.com/paper/1611.07725/) (2016)
- [Learn to Explain: Multimodal Reasoning via Thought Chains for Science Question Answering](https://mlsanity.com/paper/2209.09513/) (2022)
- [Towards VQA Models That Can Read](https://mlsanity.com/paper/1904.08920/) (2019)
- [ChartQA: A Benchmark for Question Answering about Charts with Visual and Logical Reasoning](https://mlsanity.com/paper/2203.10244/) (2022)
- [Transformer Feed-Forward Layers Are Key-Value Memories](https://mlsanity.com/paper/2012.14913/) (2020)
- [DocVQA: A Dataset for VQA on Document Images](https://mlsanity.com/paper/2007.00398/) (2020)
- [Continual Learning with Deep Generative Replay](https://mlsanity.com/paper/1705.08690/) (2017)
- [Efficient Lifelong Learning with A-GEM](https://mlsanity.com/paper/1812.00420/) (2018)
- [What Does BERT Look At? An Analysis of BERT's Attention](https://mlsanity.com/paper/1906.04341/) (2019)
- [In-context Learning and Induction Heads](https://mlsanity.com/paper/2209.11895/) (2022)
- [SEED-Bench: Benchmarking Multimodal LLMs with Generative Comprehension](https://mlsanity.com/paper/2307.16125/) (2023)
- [Are We on the Right Way for Evaluating Large Vision-Language Models?](https://mlsanity.com/paper/2403.20330/) (2024)
- [Analyzing Multi-Head Self-Attention: Specialized Heads Do the Heavy Lifting, the Rest Can Be Pruned](https://mlsanity.com/paper/1905.09418/) (2019)
- [A Diagram Is Worth A Dozen Images](https://mlsanity.com/paper/1603.07396/) (2016)
- [Robust fine-tuning of zero-shot models](https://mlsanity.com/paper/2109.01903/) (2021)
- [MaPLe: Multi-modal Prompt Learning](https://mlsanity.com/paper/2210.03117/) (2022)
- [An Empirical Study of Catastrophic Forgetting in Large Language Models During Continual Fine-tuning](https://mlsanity.com/paper/2308.08747/) (2023)
- [PathVQA: 30000+ Questions for Medical Visual Question Answering](https://mlsanity.com/paper/2003.10286/) (2020)
- [CODA-Prompt: COntinual Decomposed Attention-based Prompting for Rehearsal-Free Continual Learning](https://mlsanity.com/paper/2211.13218/) (2022)
- [InfographicVQA](https://mlsanity.com/paper/2104.12756/) (2021)
- [Expert Gate: Lifelong Learning with a Network of Experts](https://mlsanity.com/paper/1611.06194/) (2016)
- [LMMs-Eval: Reality Check on the Evaluation of Large Multimodal Models](https://mlsanity.com/paper/2407.12772/) (2024)
- [Molmo and PixMo: Open Weights and Open Data for State-of-the-Art Vision-Language Models](https://mlsanity.com/paper/2409.17146/) (2024)
- [LoRA Learns Less and Forgets Less](https://mlsanity.com/paper/2405.09673/) (2024)

## Built on by (most cited first, 11 of 11)

- [SAME: Stabilized Mixture-of-Experts for Multimodal Continual Instruction Tuning](https://mlsanity.com/paper/2602.01990/) (2026)
- [Why Fine-Tuning Encourages Hallucinations and How to Fix It](https://mlsanity.com/paper/2604.15574/) (2026)
- [Active View Selection with Perturbed Gaussian Ensemble for Tomographic Reconstruction](https://mlsanity.com/paper/2603.06852/) (2026)
- [AC3S: Adaptive Conditioning for 3D-Aware Synthetic Data Generation](https://mlsanity.com/paper/2606.31204/) (2026)
- [Sparse Subnetwork Enhancement for Underrepresented Languages in Large Language Models](https://mlsanity.com/paper/2510.13580/) (2025)
- [DocAtlas: Multilingual Document Understanding Across 80+ Languages](https://mlsanity.com/paper/2605.12623/) (2026)
- [Behemoth: Benchmarking Unlearning in LLMs Using Fully Synthetic Data](https://mlsanity.com/paper/2601.23153/) (2026)
- [CRAM: Centroid-Routing and Adaptive MoE for Multimodal Continual Instruction Tuning](https://mlsanity.com/paper/2606.02502/) (2026)
- [ProtoAda: Prototype-Guided Adaptive Adapter Expansion and Geometric Consolidation for Multimodal Continual Instruction Tuning](https://mlsanity.com/paper/2606.02576/) (2026)
- [AlphaWiSE: Adaptive Weight Interpolation for Continual Multimodal Representation Learning](https://mlsanity.com/paper/2607.15094/) (2026)
- [Render to Reason: Novel-View Semantic Prediction Improves Spatial Understanding in VLMs](https://mlsanity.com/paper/2610.05417/) (2026)

## Abstract (arXiv)

How can we teach large multimodal models (LMMs) new skills without erasing prior abilities? We study sequential fine-tuning on five target skills while monitoring general ability on eight held-out benchmarks across three model families. Surprisingly, we find that performance lost on held-out tasks after fine-tuning on one skill can partly recover when the model is subsequently tuned on a different skill. We trace this behavior to a measurable shift in the output token distribution, manifested through a simple counting-bias probe that shows the shift co-varies with forgetting. Guided by this insight, we identify two simple, robust tuning recipes that learn strongly while limiting drift: (i) updating only the self-attention projection layers (SA Proj., Delta learning +24.9 / Delta held-out forgetting -0.6), and (ii) updating only the MLP Gate&Up while freezing the Down projection (+30.5 / -2.1). Both substantially outperform full-LLM tuning (+31.8 / -23.3) in the learning-forgetting trade-off. We also compare against common forgetting mitigation methods: Learning without Forgetting (LwF), LoRA, Mixture-of-Experts, and weight-space interpolation (WiSE-FT), and find that our selective tuning recipes match or exceed their learning-stability balance while remaining simpler, requiring no replay, auxiliary parameters, or per-stage tuning. These results hold across LLaVA-OneVision, LLaVA-NeXT, and Qwen2.5-VL, confirming that the key to teaching LMMs new skills without forgetting lies in controlling output distribution shift by choosing which components to tune. Code will be made available.